Ability-to-Repay Under Reg Z 1026.43 in AI-Assisted Underwriting: The Eight Factors, the Revised General QM Price Test, and the Documentation Boundary the Agent Cannot Cross Alone

The Rule the Origination Program Cannot Get Wrong

The Ability-to-Repay rule at 12 CFR 1026.43 is the closest thing consumer-mortgage origination has to a foundational rule. The statute at 15 USC 1639c requires creditors to make a reasonable and good-faith determination based on verified and documented information that the consumer will have a reasonable ability to repay a covered residential mortgage loan. The rule implements the statute with the eight-factor test at 1026.43(c), the Qualified Mortgage safe-harbor and rebuttable-presumption structure at 1026.43(e), and the documentation verification requirements at 1026.43(c)(4).

The consequences of getting the rule wrong are the largest of any consumer-mortgage rule. A loan that fails the ATR analysis and does not qualify for the QM safe harbor exposes the creditor to a private action under TILA §1640(a) with a three-year statute of limitations under 15 USC 1640(e) reaching actual damages, enhanced statutory damages for ATR violations, court costs, and attorney fees. Independently, TILA §1640(k) allows the consumer to raise an ATR violation as a defense by recoupment or setoff in any judicial or nonjudicial foreclosure or other action to collect the debt, without regard to the private-action limitations period. Section 1640(k) means the creditor, assignee, and holder exposure does not close at year three; a loan originated in violation of the ATR rule can be defended against in a foreclosure filed years later. Assignee derivative liability runs through 15 USC 1641, and CFPB enforcement runs under the Bureau's UDAAP and TILA authorities. The consequence of the wrong ATR determination is a loan whose value in the market is impaired for years and whose enforceability against a defending borrower is at issue for the loan's life.

We build the agent that participates in mortgage-origination intake and underwriting on lender platforms. The architecture below is what we run to keep the agent's contribution to the ATR analysis inside the boundaries the rule sets and to produce the documentation the QM safe harbor and the non-QM defense both depend on.

The Eight Factors and What the Agent Actually Computes

The eight factors at 1026.43(c)(2) are the reasonable-ability-to-repay determination's substantive content. The creditor considers, at a minimum: the consumer's current or reasonably expected income or assets other than the value of the dwelling securing the loan; the consumer's current employment status; the monthly payment on the covered transaction; the monthly payment on any simultaneous loan the creditor knows or has reason to know will be made; the monthly payment for mortgage-related obligations (property taxes, insurance, HOA dues, ground rent); the consumer's current debt obligations, alimony, and child support; the consumer's monthly debt-to-income ratio or residual income; and the consumer's credit history.

The eight factors are not a formula. The rule at 1026.43(c)(1) says the determination is a reasonable and good-faith one based on verified and documented information, and the factors are the minimum content of the determination. The creditor's actual analysis can weight the factors differently for different consumer profiles, can consider additional factors not enumerated, and can accept a specific factor's weakness offset by another factor's strength.

The agent's role is to produce the factor-by-factor computation with the verified inputs, to identify the factor pattern the consumer's profile produces, and to present the computation to the underwriter for the reasonable-and-good-faith determination. The agent does not conclude the ATR analysis; the underwriter does. The agent's contribution is the accuracy and completeness of the computation the underwriter is deciding on, and the agent's operational value is that the computation the underwriter sees is the correct one.

The Third-Party Verification the Rule Requires and the Documentation Architecture

The rule at 1026.43(c)(4) requires the creditor to verify the consumer's income and assets using reliable third-party records. The rule specifies documentation types: W-2s, tax returns and IRS transcripts, payroll statements, financial-institution records, records from the consumer's employer, records from a government agency, records from a third party the consumer authorized to release information to the creditor. The rule does not accept the consumer's oral statement of income or assets as verification; the third-party record is the standard.

The verification of employment status at 1026.43(c)(4) similarly requires third-party records. Written verification of employment, a payroll statement showing current status, or equivalent third-party sources are the acceptable methods. The consumer's statement of employment status is not sufficient absent the third-party corroboration.

The verification of debts, alimony, and child support runs through the credit report the creditor pulls under 1026.43(c)(4)(iv), through court orders for alimony and child support, and through the consumer's disclosure of debts not appearing on the credit report with the creditor's verification of the disclosure. The creditor's file has to show the specific documents the creditor relied on for each factor's verification.

The agent's document intake collects the third-party documents from the consumer through secure channels, verifies each document's authenticity against the source (payroll integration with the employer's payroll provider, IRS transcript through the consumer-authorized 4506-C process, credit-report pull from the consumer-reporting agency), and produces a document-by-document record of the specific source and the specific data extracted. The record is the audit file for the verification, and the file's structure allows the examiner or the consumer's defense counsel to trace each factor's computation back to the specific third-party record it was based on.

The Revised General QM and the APR-to-APOR Price Test

The Qualified Mortgage safe harbor at 1026.43(e) is the creditor's protection against ATR liability, and the QM definitions have been through the most-revised part of the rule. The current General QM definition at 1026.43(e)(2) as amended by the CFPB's 2020 final rule removed the 43-percent DTI cap and the Appendix Q verification requirements that had governed since 2014, and replaced them with a price-based threshold: a General QM is a loan whose annual percentage rate does not exceed the Average Prime Offer Rate for a comparable transaction by more than a specified amount at consummation.

The eligibility threshold at 1026.43(e)(2)(vi) is 2.25 percentage points for a first-lien covered transaction with a loan amount at or above an indexed threshold, 3.5 percentage points for a first-lien loan in the middle size band and for subordinate-lien loans at or above the indexed threshold, and 6.5 percentage points for the smallest first-lien loans, for subordinate-lien loans below the indexed threshold, and for first-lien covered transactions secured by a manufactured home below the smaller-loan threshold. A loan whose APR exceeds APOR by more than the applicable stepped threshold is not a General QM. The APOR is published weekly by the Federal Financial Institutions Examination Council, and the creditor's determination uses the APOR in effect at the time the interest rate is set (rate-lock date under 1026.43(e)(2)(vi)).

Inside the General QM population, the safe harbor at 1026.43(e)(1)(i) applies to loans that are not "higher-priced" as defined at 1026.43(b)(4): the safe harbor cutoff for a first-lien covered transaction is APOR plus 1.5 percentage points, with parallel cutoffs for subordinate liens and manufactured-home loans. A General QM whose APR sits between the safe-harbor cutoff and the General QM eligibility cutoff (APOR + 1.5 to APOR + 2.25 for a large first-lien loan) is a higher-priced QM with a rebuttable presumption of compliance rather than the full safe harbor. The 1.5-point cutoff and the 2.25-point cutoff serve different functions, and treating either one as the General QM eligibility line would misclassify the loan.

The agent's ATR analysis produces the General QM determination as one of the analysis's outputs, computes the APR-to-APOR spread with the specific rate-lock date and APOR reference, and identifies whether the loan qualifies for the safe harbor or the rebuttable presumption. The determination and the underlying computation are stored in the file with the specific APOR reference, the specific APR at consummation, and the specific loan-amount threshold that applied.

The Loan Feature Requirements the General QM Still Imposes

Beyond the price test, the General QM at 1026.43(e)(2) imposes specific product-feature requirements. The loan cannot have negative amortization, interest-only payments, or a balloon payment. The loan term cannot exceed 30 years. The total points and fees cannot exceed 3 percent of the loan amount for larger loans, with stepped thresholds for smaller loans. The underwriting has to consider the consumer's monthly debt-to-income ratio at the maximum interest rate that will apply during the first five years and the fully amortizing scheduled payment.

The product-feature requirements are largely absent from the price test but they gate access to the safe harbor independently. A loan that meets the price test but has a product feature outside the QM allowed set is not a General QM regardless of the price, and the ATR analysis has to proceed as a non-QM analysis with the eight-factor test as the substantive standard.

The agent's product-feature check runs at intake and identifies any product feature that would exclude the loan from the QM safe harbor. The check is deterministic against the rule's specific feature list, and the check's output routes the loan into the QM workflow or the non-QM workflow with the specific reasoning documented.

The Small Creditor QM and the Seasoned QM Layers

The rule provides additional QM categories for specific creditor and loan populations. The Small Creditor QM at 1026.43(e)(5) is available for creditors that meet the small-creditor definition at 1026.35(b)(2)(iii) and that originate the loan intending to hold it in portfolio. The Small Creditor QM has a higher APR-to-APOR threshold than the General QM (3.5 percentage points rather than 1.5) and different balloon-payment allowances.

The Seasoned QM at 1026.43(e)(7) is a category the CFPB added in 2020 for loans that have been held in portfolio for at least 36 months with an acceptable payment history and that were originated as non-QM or as higher-priced QMs. The Seasoned QM's addition allows the loan to qualify for the safe harbor after the seasoning period without having qualified at origination.

The agent's QM classification identifies which QM category the loan is a candidate for, applies the specific requirements for that category, and documents the classification. A loan that qualifies as a Small Creditor QM has a documentation set that includes the creditor's small-creditor status verification and the loan's portfolio-hold intention documented in the origination file. A loan that will be a candidate for Seasoned QM classification later is flagged for the servicing system to track the seasoning period and payment history.

The Non-QM Analysis and the Eight-Factor Documentation That Defends It

A loan that does not qualify for any QM category is a non-QM loan, and the creditor's ATR analysis proceeds as the eight-factor analysis without the QM's safe-harbor protection. Non-QM loans are legitimate and commercially important for consumers whose profile does not fit the QM categories (self-employed borrowers with complex income, high-DTI-but-high-asset borrowers, jumbo loans in some pricing environments). The non-QM market is a meaningful share of the origination universe, and the ATR analysis on non-QM loans is the substantive protection the creditor and any assignee have.

The eight-factor documentation on a non-QM loan is substantially more extensive than the QM documentation. Each of the eight factors has to be verified with third-party documentation, the reasonable-and-good-faith determination has to reflect the specific factors' consideration, and the underwriter's judgment on how the factors combined into a reasonable ability-to-repay determination has to be documented in a form the future consumer defense or examiner review can follow.

The agent's non-QM analysis produces the eight-factor computation with the verification for each factor, the identification of the specific ATR risk drivers in the consumer's profile, and the presentation to the underwriter with the reasoning the underwriter can accept, adjust, or reject. The underwriter's decision is documented with the specific reasoning applied to each factor and to the overall determination. A non-QM loan's underwriting file is longer and more explicitly-reasoned than a QM loan's file, because the safe-harbor protection is not there and the substantive analysis has to carry the defense.

The Simultaneous-Loan Factor and the Piggyback Coordination

The fourth factor at 1026.43(c)(2)(iv) is the monthly payment on any simultaneous loan the creditor knows or has reason to know will be made. The most common simultaneous-loan pattern is the piggyback second lien that reduces the first lien's loan-to-value ratio below the private-mortgage-insurance threshold, or the home-equity line of credit opened at closing to fund down payment.

The creditor's ATR analysis has to include the simultaneous loan's monthly payment in the consumer's overall debt calculation, and the agent's intake has to know whether a simultaneous loan is being originated (either by the same creditor or by a coordinating third-party creditor) and what the simultaneous loan's monthly payment will be. The agent's coordination with the piggyback lender or with the HELOC origination system produces the payment information for the ATR analysis.

The failure mode we see is the ATR analysis on a first lien that considers only the first-lien payment and misses the piggyback second lien's payment. The consumer's DTI as analyzed is understated, the ATR determination is based on an inaccurate debt figure, and the loan's ATR compliance is technically deficient. The agent's simultaneous-loan check runs on every loan and requires an affirmative answer to "is there any simultaneous loan I need to include in this analysis," with the documentation of the answer and the source of the answer in the file.

The Mortgage-Related Obligations Factor and the Escrow Consistency

The fifth factor at 1026.43(c)(2)(v) is the monthly payment for mortgage-related obligations, which includes property taxes, insurance premiums, HOA dues, and ground rent. The mortgage-related obligations figure feeds the consumer's DTI computation and also feeds the escrow analysis under Reg X 1024.17 that the servicer will run on the escrow account.

The consistency between the ATR analysis's mortgage-related-obligations figure and the escrow analysis's disbursement projections is a specific control the agent maintains. A loan whose ATR analysis estimated $4,800 in annual property taxes and whose initial escrow statement projects $6,200 has an inconsistency that either the ATR analysis was based on stale information or the escrow analysis is projecting a different figure than the ATR analysis assumed. The inconsistency's resolution goes back to the ATR analysis's file: what did the analysis rely on, and does the actual current tax amount change the reasonable-ability-to-repay determination.

The agent's coordination between the origination-side ATR analysis and the servicing-side escrow analysis is that both analyses use the same source for the property-tax figure (typically the county assessor's record or a comparable authoritative source), and any change in the source between origination and servicing is flagged as a potential re-analysis event. The file's consistency across the origination-servicing boundary is a specific artifact the examiner will look for.

The Interaction With TRID Timing and the AI Underwriting Cycle

The ATR analysis has to conclude before consummation, but the analysis's key elements can shift during the loan's timeline as documentation is received and verification progresses. The TRID timing rules for the Loan Estimate and Closing Disclosure have their own cadence, and the ATR analysis's completion has to align with the TRID timing without creating a race between the two.

The pattern we run is that the agent's ATR analysis produces a preliminary determination early in the loan's cycle (typically at the initial Loan Estimate), a mid-cycle determination as documentation is received, and the final determination in advance of the Closing Disclosure delivery so that the CD accurately reflects the terms the ATR analysis relies on. The final determination is the underwriter's decision, and the underwriter's decision has to be in the file before the CD is issued.

A change in the loan's terms after the CD but before consummation triggers a re-verification of the ATR analysis's dependent factors and, if the change is material, a re-issuance of the CD with the corresponding TRID timing consequences. The agent's coordination across the ATR analysis and the TRID cadence is a specific operational discipline the underwriting workflow requires.

The Fair-Lending Interaction and the Disparate-Impact Analysis

The ATR analysis's eight factors are neutral on their face and are applied consistently to all applicants under the rule. The consistency does not by itself immunize the analysis from disparate-impact scrutiny under ECOA and the fair-lending framework the CFPB and the DOJ pursue. A factor's weighting, a documentation-verification standard, or a QM-vs-non-QM classification pattern that produces different outcomes for protected classes triggers the disparate-impact analysis regardless of the analysis's per-loan compliance with 1026.43.

The agent's ATR contribution has to be evaluated at the population level for disparate-impact patterns, and the model risk management framework we described at SR 11-7 and the NIST AI RMF applies. The population-level review of ATR outcomes by protected class is a specific analysis the fair-lending function runs at least annually, and the review's findings feed the ATR analysis's calibration for the next cycle.

The Adverse-Action Interaction When the Answer Is No

A loan that fails the ATR analysis is a loan the creditor cannot make. The consumer's application receives an adverse-action notice under the Regulation B ECOA and the FCRA rules we described separately, with the specific reasons for the denial in the notice's reason codes. The ATR-related reasons include income insufficiency to support the requested payment, employment status not verified as required, debt-to-income ratio above the acceptable range, credit history indicating high risk of default, or the specific combination of factors that produced the reasonable-good-faith determination against the loan.

The adverse-action notice's specific reasons have to reflect the actual ATR analysis's specific weaknesses, not generic denial reasons. A consumer whose denial notice says "insufficient income" is a consumer who can respond specifically; a consumer whose denial notice says "does not meet our underwriting requirements" is a consumer who cannot respond and whose adverse-action notice violates the specificity requirement at Reg B 1002.9.

The agent's contribution to the adverse-action notice is to translate the ATR analysis's specific findings into the adverse-action notice's reason codes with the specificity the rule requires. The notice's content is generated from the analysis's data rather than from a generic denial template, and the notice serves the consumer's understanding of what the denial was based on and the consumer's opportunity to address the specific weaknesses in a future application.

The Documentation Retention the Assignee Will Ask For

The ATR analysis's file has to be maintained for at least three years under 1026.25, and the assignee's liability under 1026.43(l) means the assignee will want the analysis's file at the loan's sale. A loan sold into the secondary market with an incomplete or ambiguous ATR file is a loan whose assignee's due diligence will find the ambiguity and either reduce the loan's price, decline the purchase, or require the originator to remedy the file's deficiencies before purchase.

The agent's file production produces the ATR analysis in a form the secondary-market due diligence can review at scale, with each factor's verification documents, the QM classification and its supporting computation, the underwriter's determination and reasoning, and any adverse-action notice or approval documentation. The file's structure is standardized across the originator's loan population, and the secondary-market buyer's diligence can process the population efficiently.

The Audit File the ATR Analysis Produces Per Loan

The artifact set per loan that the regulator, the assignee, or the consumer's defense counsel will ask for includes the eight-factor computation with the specific values and the source documents for each; the QM classification with the price-test computation, the product-feature check, and the specific QM category; the underwriter's determination with the reasoning applied to the specific factors and the overall analysis; the adverse-action notice or the approval documentation with the specific reasons or terms; the CD's consistency check against the ATR analysis; and the file's audit history showing any changes made during the loan's cycle with the timestamps and the approvals.

The file's completeness supports the safe-harbor claim on QMs, the substantive-defense on non-QMs, the disparate-impact review on the population, and the assignee's due diligence on secondary-market sales. The file's incompleteness is the single largest risk to the ATR compliance posture, and the agent's file-production discipline is the operational lever that keeps the file complete.

The Failure Mode We Engineer Against

The pattern that produces the worst ATR outcomes is the origination program that treats the ATR analysis as an underwriting-team function separate from the intake and documentation workflow, with the underwriter running the analysis on the file the intake team assembled and the intake team not aware of what the analysis needed to see. The file arrives at underwriting with gaps that the underwriter has to chase, the loan's timeline extends while the gaps are filled, and the pressure to close on time produces documentation shortcuts that the file's audit trail cannot fully support.

The architecture we run is that the ATR analysis is the intake's target, not the underwriter's job to backfill. The agent's intake collects the documents and produces the factor-by-factor computation as the intake proceeds, with the underwriter's role being to review the completed analysis and to make the reasonable-and-good-faith determination. The gaps that would have been the underwriter's to chase are addressed at intake with the consumer available to provide what is missing, and the underwriter's file is complete before the underwriting review begins.

The consequence is that underwriting time per loan reduces meaningfully because the underwriter is deciding rather than assembling, the loan's cycle time shortens, and the file's audit posture is stronger because the assembly happened when the consumer was available and the third-party sources were fresh.

The Honest Read

The ATR/QM rule is dense, has been revised repeatedly, and has real teeth. The rule's compliance is the foundation of the origination program's legal defensibility, and the AI agent's contribution to the compliance is the accuracy and completeness of the underlying analysis the underwriter's reasonable-and-good-faith determination rests on. The creditors whose ATR programs are strong are the creditors whose secondary-market economics are strong, whose fair-lending posture is defensible, and whose eventual defense in a consumer action is anchored in a file that supports the rule's specific requirements.

The AI agent does not conclude the ATR analysis. The rule allocates the determination to the creditor, and the creditor's designated underwriter carries the reasonable-and-good-faith judgment. The agent's role is to produce the computation the underwriter needs, to verify the documentation the rule requires, to identify the QM classification and the safe-harbor status, and to keep the file in the form the eventual review will require. The judgment is the human's; the operational discipline that supports the judgment is the agent's.

We have written separately on the adverse-action notice rules under ECOA/Reg B, on the fair-lending disparate-impact framework, on the model-risk-management architecture, on the TRID timing rules, on the HMDA data-integrity framework, and on the Reg X escrow analysis that runs the mortgage-related-obligations figure into servicing. The ATR analysis sits at the center of the origination program and connects to each of these rules; the agent that runs across all of them with the same discipline is the agent whose contribution to the origination program compounds across the loan's lifecycle.